The Physics of Artificial Intelligence (PAI)The Defense Advanced Research Projects Agency (DARPA) Defense Sciences Office (DSO) is issuing a Disruption Opportunity (DO) Special Notice (SN) inviting submissions of innovative basic research concepts exploring radically new architectures and approaches in Artificial Intelligence (AI) that incorporate prior knowledge, such as known physical laws, to augment sparse data and to ensure robust operation. This DO SN is part of DARPA’s portfolio of ongoing AI research and is issued under the Program Announcement for Disruptioneering, DARPA-PA-18-01. All proposals in response to the technical area(s) described herein will be submitted to DARPA-PA-18-01 and if selected, will result in an award of an Other Transaction (OT) for prototype project not to exceed $1,000,000.Defense Advanced Research Projects AgencyDARPA_46ad30a4-ad5d-11df-9c96-10167a64ea2aFor more than five decades, DARPA has been a leader in spurring groundbreaking research and development (R&D) that facilitated the advancement and application of “First Wave” (rule based) and “Second Wave” (statistical learning based) AI technologies. Today, DARPA continues to lead innovation in AI research as it continues to fund a broad portfolio of R&D programs, ranging from basic research to advanced technology development, which will help shape a future for AI technology where machines may serve as trusted and collaborative partners in solving problems of importance to national security. DARPA believes this future will be realized upon the development and application of “Third Wave” AI technologies, where systems are capable of acquiring new knowledge through generative contextual and explanatory models. The Physics of AI (PAI) basic research Disruption Opportunity supports this vision. DODIt is anticipated that AI will play an ever larger role in future Department of Defense (DoD) activities, ranging from scientific discovery, to human-machine collaboration, to real-time sensor processing, to the control and coordination of a variety of distributed, intelligent and autonomous composable systems. However, despite rapid and accelerating progress of AI in the commercial sector -- particularly in the subfield of machine learning -- AI has not yet been successfully integrated into the most transformative DoD applications, for reasons that have included [factors documented as values in this StratML rendition]Systems are capable of acquiring new knowledge through generative contextual and explanatory models_34974d0a-846e-11e8-8d18-276ee53a5cccTo develop novel AI architectures, algorithms and approaches that "bake in" the physics, mathematics and prior knowledge relevant to an application domain in order to address the technical challenges in application of AI in scientific discovery, human-AI collaboration, and a variety of defense applications._34974e0e-846e-11e8-8d18-276ee53a5cccTransformation... despite rapid and accelerating progress of AI in the commercial sector -- particularly in the subfield of machine learning -- AI has not yet been successfully integrated into the most transformative DoD applications, for reasons that have included:TrustThe demanding levels of trust, safety and performance guarantee required of AI systems in defense applicationsSafetyPredictionThe lack of success of deep learning constructs in causal, predictive modeling of complex nonlinear dynamic systemsLearningThe acknowledged difficulties of machine learning architectures and training protocols in dealing with incomplete, sparse and noisy dataRobustnessThe lack of robustness, which makes AI image recognition systems potentially subject to a variety of adversarial spoofingOpennessThe inherent challenges faced by AI approaches in dealing with "Open World problems", e.g., in unstructured environments with unknown and hidden states, as compared to relatively well-structured application domains (e.g. games) where the system state is fully observable and interaction rules are knownPerformanceThe difficulty in obtaining useful performance guarantees and limits or even to know what questions can be asked of an AI system and whether the answers make senseInnovationPAI is seeking innovative approaches that address the [goal and objectives] and can substantially improve upon current machine learning approaches in bringing "deep insight" into physics-centric application domains. AI architectures, algorithms and approaches that make use of DNNs as one of several component are welcome, but conventional learning algorithms using DNNs (including convolutional and recurrent neural networks) by themselves are not considered likely to meet the broad goals of the program.InsightHybridizationHybrid architectures are encouraged that embed hierarchical physical models into generative cores; that incorporate manifold learning techniques; that incorporate operator theoretic spectral methods, and/or that bake in topological knowledge, group symmetries, projection knowledge, or gauge invariances into the network architecture.ResilienceGenerative approaches that can reproduce the multiscale structures of observed data; distinguish between semantic and stylistic differences; are resilient to noise, data dropouts, data biases, and adversarial spoofing; and can learn with minimal labeled data are also encouraged.Physics AwarenessEmbed physics and prior knowledge into AI in a principled way to augment sparse data and to move from correlation to generative models that are causal and explanative_34974e9a-846e-11e8-8d18-276ee53a5cccMultidisciplinary TeamsIt is anticipated that multidisciplinary teams incorporating domain expertise, physics, mathematics, AI, statistics, information theory, control theory and additional disciplines will be required to achieve the transformational goals of the program.PhysicistsMathematiciansAI ExpertsStatisticiansInformation TheoristsControl Theorists("put the physics into AI"). Demonstrate these “physics-aware” models in DoD-relevant applications... Data-driven machine learning techniques have proven successful in leveraging massive training data to answer questions narrowly registered around the initial training set and questions. Deep artificial neural networks (DNNs) are extremely expressive in approximating arbitrary nonlinear functions, extracting features from data, and producing useful reduced-dimensional representations for classification purposes. Advanced computational platforms now enable the training of hundred-layer-deep networks using backpropagation methods that encompass hundreds of thousands to millions of parameters (the weights of the DNNs) as long as sufficient training data exist. However, despite some successes in transfer learning and one-shot learning, it has proven difficult for DNNs to generalize beyond their initial set of training questions. In general DNN’s are not generative, although generative models such as variational autoencoders (VAEs), generative adversarial networks (GANs) and hybrid models exist and have been employed in specialized domains. PAI hypothesizes that many challenges associated with current state-of-the art machine learning and DNN-based AI can be overcome, especially in physically-grounded application domains, by baking in physics from the outset. Here, "putting the physics into AI" broadly means effectively exploiting a diverse set of prior knowledge, including scientific and mathematical knowledge relevant to the problem at hand. Proposals should address the following research project objectives:PrototypeDevelop an AI prototype that makes optimal use of both observational and experimental data, simulated data, and prior knowledge._34974f26-846e-11e8-8d18-276ee53a5ccc1Architectures and algorithms incorporating reusable building blocks that need not be retrained for every narrow domain or for every new set of questions are preferred. AI architectures and algorithms should embed or bake in prior knowledge, including scientific knowledge, mathematical/topological knowledge, statistical models, logical inference, linguistic or physical grammars, symmetries, conservation laws, and other physical constraints in order to overcome the limitations of sparse, noisy or incomplete data and in order to learn resilient and parsimonious representations that are physically meaningful. The aim is to learn the structure of underlying generating functions, grammars, or relationships, not simply classify the structures found in observed data.DemonstrationDemonstrate an AI prototype system using simulated and/or real data in a representative DoD-relevant systems application._34974fee-846e-11e8-8d18-276ee53a5ccc2Examples of relevant application domains include predictive nonlinear systems control; satellite or radar image processing; high dimensional systems modeling; and the use of AI in human-machine collaborative scientific discovery and exploitation.LimitationsAddress the fundamental performance limits of your prototype system. _3497508e-846e-11e8-8d18-276ee53a5ccc3Specifically, describe your approach in terms of its accuracy, generalizability (ability to effectively predict behaviors outside of the training set or to perform in other domains), robustness (to noise, sparse data and adversarial spoofing), and data and computational requirements. Quantitatively compare your physics-aware approach with current practice.2018-07-062018-08-062018-07-10https://www.fbo.gov/index?s=opportunity&mode=form&id=9efd2ae84ab73e1dfcbb96b4fbd802f8&tab=core&_cview=0OwenAmburOwen.Ambur@verizon.net